Because of their high ground resolution and their ability to provide stereoscopic observation, Satellite Pour l'Observation de la Terre (SPOT) images have been widely used in the field of land-use observation, agriculture detection, forestry-resource surveying, environmental monitoring and large-scale photogrammetric mapping. However, SPOT images have no blue channel. It is therefore difficult for users to create a composite natural-colour image, which has restricted the field of application for SPOT images in areas such as fly-through of draped terrain, visual interpretation or display generation for non-remote-sensing professionals. To overcome this limitation, this article proposes a new approach for generating pseudo-natural-colour composite (NCC) representations from false-colour composite (FCC) images based on spectrum machine learning (SML). Taking samples in a spectral library as an a priori knowledge database, this article uses a machine-learning method to establish an implicit non-linear relationship between the blue band and other bands (green, red, near-infrared (NIR) and shortwave infrared (SWIR)) using a support vector machine. Then, the non-linear relationship is used on a SPOT image to simulate a new blue band. The blue band, along with the green and red bands, provides a near-true or ‘natural’ colour on the display. Experimental results show that the method is valid. The proposed ‘natural colour generator’ can be used to change false-colour images to natural-colour images. The quality of the generated pseudo-natural-colour (PNC) images is fully acceptable for manual mapping. In addition, the method can also be applied to other satellite images to simulate natural-colour images.